一类新的非单调记忆梯度法及其全局收敛性(英文)

A New Class of Nonmonotone Memory Gradient Method and Its Global Convergence

在非单调Armijo线搜索的基础上提出一种新的非单调线搜索,研究了一类在该线搜索下的记忆梯度法,在较弱条件下证明了其全局收敛性。与非单调Armijo线搜索相比,新的非单调线搜索在每次迭代时可以产生更大的步长,从而使目标函数值充分下降,降低算法的计算量。

Based on nonmonotone Armijo line search,the paper proposes a new nonmonotone line search and investigates a memory gradient method with this line search. Its global convergence is also proved under some mild conditions. As compared with nonmonotone Armijo rule, the new nonmonotone line search can effectively reduce the function evaluations by choosing a larger accepted stepsize at each iteration so as to reduce the computation of algorithm.